similar_index_for_filter issue about filter-pruning-geometric-median HOT 2 CLOSED

Please refer to my reply in the other issue.

Assume there are 16 filters.

1. "filter_large_index" and “filter_small_index” is for norm-based criterion, assume 3 filters selected from 16 filters. Now we have 13 remaining filters.
2. "similar_large_index" and "similar_small_index" is for FPGM criterion, assume 4 filters selected from 13 filters. Now we have 9 remaining filters.

This function ( similar_index_for_filter = [filter_large_index[i] for i in similar_small_index] ) is about mapping the "9 in 13 index" to "13 in 16 index". Details is shown in the image.

from filter-pruning-geometric-median.

Please refer to my reply in the other issue.

Assume there are 16 filters.

1. "filter_large_index" and “filter_small_index” is for norm-based criterion, assume 3 filters selected from 16 filters. Now we have 13 remaining filters.
2. "similar_large_index" and "similar_small_index" is for FPGM criterion, assume 4 filters selected from 13 filters. Now we have 9 remaining filters.

This function ( similar_index_for_filter = [filter_large_index[i] for i in similar_small_index] ) is about mapping the "9 in 13 index" to "13 in 16 index". Details is shown in the image.

Thanks!. Got it. the similar_filter is calculated based on norm filtered weights.

from filter-pruning-geometric-median.